The nonpartisan group Protect Democracy has sued four US federal agencies, demanding disclosure of the secret framework the Trump administration uses to conduct pre-release safety reviews of frontier AI models. The lawsuit alleges that almost no details about the process have been shared with the public or Congress, meaning it is unclear how models are assessed, which companies help shape the criteria, or what legal authority officials have to run the reviews at all. This matters because decisions about which AI systems are approved for release could be one of the most consequential policy areas of the administration, yet they are reportedly happening with no outside oversight from Congress, the public or independent experts.

The lawsuit names officials from the Office of the National Cyber Director, the Office of Science and Technology Policy, the Treasury Department and the Commerce Department, and seeks an order compelling them to hand over unclassified information—including the framework's text, participant identities and access criteria—by 30 September. Protect Democracy points to OpenAI's reported private arrangement limiting distribution of its advanced models to "government-vetted partners" as evidence of the framework's reach, and warns the secrecy could mask favouritism towards firms Trump favours, as seen when a judge ruled his blacklisting of Anthropic unlawful. Democratic lawmakers, including Representative Greg Casar and California state senator Josh Becker, have backed calls for transparency, especially given recent AI-related cybersecurity incidents such as an OpenAI model reportedly breaching Hugging Face.